COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "let.pub" source
C00009 ENDMK
C⊗;
.require "let.pub" source
∂MEM William Massy$John McCarthy$IBM mass storage system∞

Dear Bill:

	Congratulations on your new job.  I hope you will decide
that your empire can be trimmed.

	The IBM salesman called me yesterday to tell me about SCIP's
plan to acquire an IBM mass storage system and to arrange a meeting
to discuss connection of the AI Lab and LOTS to the system.  At first
I was enthusiastic, because I have wanted access to such a system
for many years and because I think this is just what Stanford needs.
However, because of my well known worries about SCIP's efficiency
I quizzed him about projected prices, and what I heard confirmed my
worst fears.  Here are the relevant numbers as I recall them.

	1. SCIP now charges about $50.00 per megabyte month for
storage on 3350s, and
this is scheduled to go down to about $40.00 soon.

	2. The mass storage unit to be
acquired from IBM rents for about $.30 per megabyte
month.

	3. SCIP intends an initial charge of $15.00 per megabyte month.

	4. The AI Lab can extend its current 3330 imitation system
at a cost of $12,000 per 200 megabyte drive.  Dividing $12,000 by
200 gives a purchase price of $60 per megabyte.  Four year amortization
suggests an effective cost of $1.25 per month ignoring inefficiencies
that might increase costs by a factor of two.  Comparing this alternative
with SCIP's proposed price suggests that we can amortize the cost of
more imitation 3330 storage in four months relative to using SCIP.
We also retain direct access to the information.
For us to benefit from use of a public mass storage system, the price
must be below $1.00 per megabyte month.

	5. LOTS is not in as good a position as the AI Lab.  The list
price of 200 megabyte unit from D.E.C. is three times the free market
price.  This means it would take a whole year to amortize the cost
of renting mass storage from SCIP.  However, list price is not realistic,
because D.E.C. is selling us disk units at a reduced price as part of
a package deal, because they wish to discourage us from buying a control
unit that will accept imitation 3350s.  If we were to go that route,
our marginal price would be $12,000 for 317 megabytes, so that once
the hurdle has been jumped, a price of $.60 per megabyte month would
be needed to lure us to public mass storage.

	6. The Business School should be in about the same position
as LOTS.

	The costs are realistic, because none of the PDP-10 installations
would require any additional personnel to add more drives to their
present systems.  The space requirement per drive is about that of
a file cabinet.

	I cannot believe that charging 50 times the IBM rental can be
justified by the marginal costs or even the average costs of running
SCIP.  There must be an attempt to subsidize some other service that
SCIP is running at a loss.

	I believe that public mass storage is very important to Stanford
and that the IBM device is cost-effective for that purpose.  A full
IBM system, 472 billion bytes, would store the million most frequently
used books in the Stanford library at an affordable cost.  The library
could be accessible from anywhere on the campus or anywhere in the
world.

	%3I propose that Stanford solicit management proposals for
operating mass storage, the most important part being the rental
that would be charged and the amortization plan%1.  I believe that
IMSSS could make a proposal much more attractive than the current
SCIP plan.  The SCIP plan is unlikely to be cost-effective for other computer
systems on campus.  Besides IMSSS, outside vendors should be solicited.
The AI Lab or LOTS would make a proposal only if no other organization
made a reasonable proposal.

.sgn
.skip
cc: W. Miller, C. Dickens, Arjay Miller, P. Suppes, T. Rindfleisch, R. Gorin